A two-lab direct replication attempt of Southgate, Senju and Csibra (2007)

Testing theory of mind in large language models and humans

At the core of what defines us as humans is the concept of theory of mind: the ability to track other people's mental states. The recent development of large language models (LLMs) such as ChatGPT has led to intense debate about the possibility that these models exhibit behaviour that is indistinguishable from human behaviour in theory of mind tasks. Here we compare human and LLM performance on a comprehensive battery of measurements that aim to measure different theory of mind abilities, from understanding false beliefs to interpreting indirect requests and recognizing irony and faux pas. We tested two families of LLMs (GPT and LLaMA2) repeatedly against these measures and compared their performance with those from a sample of 1,907 human participants. Across the battery of theory of mind tests, we found that GPT-4 models performed at, or even sometimes above, human levels at identifying indirect requests, false beliefs and misdirection, but struggled with detecting faux pas. Faux pas, however, was the only test where LLaMA2 outperformed humans. Follow-up manipulations of the belief likelihood revealed that the superiority of LLaMA2 was illusory, possibly reflecting a bias towards attributing ignorance. By contrast, the poor performance of GPT originated from a hyperconservative approach towards committing to conclusions rather than from a genuine failure of inference. These findings not only demonstrate that LLMs exhibit behaviour that is consistent with the outputs of mentalistic inference in humans but also highlight the importance of systematic testing to ensure a non-superficial comparison between human and artificial intelligences.

Flexible social monitoring as revealed by eye movements: Spontaneous mental state updating triggered by others' unexpected actions

Successful interactions require not only representing others' mental states but also flexibly updating them, whenever one's original inferences may no longer hold. Such situations arise, for instance, when a partner's behavior is incongruent with one's expectations. Although these situations are rather common, the question whether people update others' mental states spontaneously upon encountering unexpected behaviors and whether they use the updated mental states in novel contexts, has been largely unexplored. We addressed these issues in two experiments. In each experiment participants first performed an anticipatory looking task, reacting to a virtual 'partner', who categorized pictures based on their ambiguous or non-ambiguous color. Importantly, to perform the task participants did not have to track their partner's perspective. Following a correct categorization phase, the 'partner' started to systematically miscategorize one of the ambiguous colors (e.g., as if she would now believe that the greenish blue is green). We measured how participants' anticipatory looking preceding the partner's categorization changed across trials. Afterward, we asked whether participants implicitly transferred their knowledge about the partner's updated perspective to a new task. Finally, they performed an explicit perspective-taking task, to test whether they selectively updated the partner's perspective, but not their own. Results revealed that correct anticipations started to emerge only after a few miscategorizations, indicating the spontaneous updating of the other's perspective regarding the miscategorized color. Signatures of updating emerged somewhat earlier when the partner made similarity judgments (Experiment 2), highlighting the subjective nature of her decisions, compared to when following an explicit color-categorization rule (Experiment 1). In the explicit perspective-taking task of both experiments, roughly half of the participants could categorize items according to the partner's (spontaneously updated) perspective and also used their partner's updated perspective in the implicit transfer task to some degree, while they were the ones who displayed more pronounced anticipatory patterns as well. Such data provides strong evidence that the observed changes in anticipatory looking reflect spontaneous and flexible mental state updating. In addition, the findings also point to a high individual variability both in the updating of attributed mental states and the use of the updated mental state content.

Do toddlers reason about other people's experiences of objects? A limit to early mental state reasoning

Human social life requires an understanding of the mental states of one's social partners. Two people who look at the same objects often experience them differently, as a twinkling light or a planet, a 6 or a 9, and a random cat or Cleo, their pet. Indeed, a primary purpose of communication is to share distinctive experiences of objects or events. Here, we test whether toddlers (14-15 months) are sensitive to another agent's distinctive experiences of pictures when determining the goal underlying the agent's actions in a minimally social context. We conducted nine experiments. Across seven of these experiments (n = 206), toddlers viewed either videotaped or live events in which an actor, whose perspective differed from their own, reached (i) for pictures of human faces that were upright or inverted or (ii) for pictures that depicted a rabbit or a duck at different orientations. Then either the actor or the toddler moved to a new location that aligned their perspectives, and the actor alternately reached to each of the two pictures. By comparing toddlers' looking to the latter reaches, we tested whether their goal attributions accorded with the actor's experience of the pictured objects, with their own experience of the pictured objects, or with no consistency. In no experiment did toddlers encode the actor's goal in accord with his experiences of the pictures. In contrast, in a similar experiment that manipulated the visibility of a picture rather than the experience that it elicited, toddlers (n = 32) correctly expected the actor's action to depend on what was visible and occluded to him, rather than to themselves. In a verbal version of the tasks, older children (n = 35) correctly inferred the actor's goal in both cases. These findings provide further evidence for a dissociation between two kinds of mental state reasoning: When toddlers view an actor's object-directed action under minimally social conditions, they take account of the actor's visual access to the object but not the actor's distinctive experience of the object.

This is me! Neural correlates of self-recognition in 6- to 8-month-old infants

Historically, evidence of self-recognition in development has been associated with the "rouge test"; however, this has been often criticized for providing a reductionist picture of self-conscious behavior. With two event-related potential (ERP) experiments, this study investigated the origin of self-recognition. Six- to eight-month-old infants (42 males and 35 females, predominately White, tested in the UK in 2022-2023) were presented with images of their face, another peer's face, and their mother's face (N = 38, Exp.1), and images of their own face morphed into another peer's face (N = 39, Exp.2). Results showed an enhanced P100 in infants' ERP response to their own face compared to others' faces (Exp.1 only), suggesting the presence of an enhanced attentional mechanism to one own's face as early as 6 months.

Investigating belief understanding in children in a nonverbal ambiguous displacement and communication setting

Finding ways to investigate false belief understanding nonverbally is not just important for preverbal children but also is the only way to assess theory of mind (ToM)-like abilities in nonhuman animals. In this preregistered study, we adapted the design from a previous study on pet dogs to investigate false belief understanding in children and to compare it with belief understanding of those previously tested dogs. A total of 32 preschool children (aged 5-6 years) saw the displacement of a reward and obtained nonverbal cueing of the empty container from an adult communicator holding either a true or false belief. In the false belief condition, when the communicator did not know the location of the reward, children picked the baited container, but not the cued container, more often than the empty one. In the true belief condition, when the communicator witnessed the displacement yet still cued the wrong container, children performed randomly. The children's behavior pattern was at odds with that of the dogs tested in a previous study, which picked the cued container more often when the human communicator held a false belief. In addition to species comparisons, because our task does not require verbal responses or relational sentence understanding, it can also be used in preverbal children. The children in our study behaved in line with the existing ToM literature, whereas most (but not all) dogs from the previously collected sample, although sensitive to differences between the belief conditions, deviated from the children. This difference suggests that using closely matched paradigms and experimental procedures can reveal decisive differences in belief processing between species. It also demonstrates the need for a more comprehensive exploration and direct comparison of the various aspects of false belief processing and ToM in different species to understand the evolution of social cognition.

Evaluative contexts facilitate implicit mentalizing: relation to the broader autism phenotype and mental health

One promising account for autism is implicit mentalizing difficulties. However, this account and even the existence of implicit mentalizing have been challenged because the replication results are mixed. Those unsuccessful replications may be due to the task contexts not being sufficiently evaluative. Therefore, the current study developed a more evaluative paradigm by implementing a prompt question. This was assessed in 60 non-autistic adults and compared with a non-prompt version. Additionally, parents of autistic children are thought to show a genetic liability to autistic traits and cognition and often report mental health problems, but the broader autism phenotype (BAP) is an under-researched area. Thus, we also aimed to compare 33 BAP and 26 non-BAP mothers on mentalizing abilities, autistic traits, compensation and mental health. Our results revealed that more evaluative contexts can facilitate implicit mentalizing in BAP and non-BAP populations, and thus improve task reliability and replicability. Surprisingly, BAP mothers showed better implicit mentalizing but worse mental health than non-BAP mothers, which indicates the heterogeneity in the broader autism phenotype and the need to promote BAP mothers' psychological resilience. The findings underscore the importance of contexts for implicit mentalizing and the need to profile mentalizing and mental health in BAP parents.

False belief understanding in deaf children: what are the difficulties?

Children with cochlear implants (CIs) demonstrate proficiency in verbal-story elicited-response (VS-ER) false-belief tasks, such as the Sally & Ann task, at a similar age as typically developing hearing children. However, they face challenges in non-verbal spontaneous-response (NV-SR) false-belief tasks, measured via looking times, which hearing infants typically pass by around 2 years of age, or earlier. The purpose of the present study was to examine whether these difficulties remain in a non-verbal-story elicited-response (NVS-ER) false-belief task, in which children are offered the opportunity to provide an elicited response to a non-verbal-story task. A total of thirty 4- to 8-year-old children with CI-s and hearing children completed three different kinds of false-belief tasks. The results showed that children with CI-s performed above chance level on the verbal task (i.e., VS-ER task), but not on the two non-verbal tasks, (i.e., NVS-ER and NV-SR tasks). The control group of typically developing hearing children performed above chance on all three kinds of tasks (one-tailed significance level). Our findings highlight the importance of external narrative support for children with CIs in tasks that involve mental perspective-taking, and specifically predicting actions based on false beliefs.

How robust are egocentric and altercentric interference effects in social cognition? a test with explicit and implicit versions of a continuous false belief task

It has been long assumed that meta-representational theory of mind (ToM) -our ability to ascribe mental states to ourselves and other people- emerges around age four as indicated in performance on explicit verbal false belief tasks. In contrast, newer studies assessing false belief understanding with implicit, non-verbal measures suggest that some form of ToM may be present even in infancy. But these studies now face replication issues, and it remains unclear whether they can provide robust evidence for implicit ToM. One line of research on implicit ToM, however, may remain promising: Studies that tap so-called altercentric biases. Such biases occur when agents in their judgments about the world are influenced (perform slower, more error-prone) in light of another agent's deviating perspective even if that perspective is completely irrelevant to the task; they thus can be seen as indicators of spontaneous and implicit ToM. Altercentric biases are the mirror images of egocentric biases (agents are influenced by their own perspective when evaluating another agent's deviating perspective). In three studies with adults, we aimed to tap both egocentric and altercentric interference effects within the same task format. We used the so-called Sandbox task, a false belief task with continuous locations. In Study 1, we tested an online adaptation of the Sandbox task, which we also used to explore potential cross-cultural differences in these biases. Studies 2 and 3 combined the Sandbox task with mouse-tracking measures. These studies revealed neither egocentric nor altercentric biases. These null results are discussed with regard to the question whether absence of evidence here may present evidence of absence of such spontaneous perspective-taking biases or merely false negatives.

The social roots of self development: from a bodily to an intellectual interpersonal dialogue

In this paper, we propose that interpersonal bodily interactions represent a fertile ground in which the bodily and psychological self is developed, gradually allowing for forms of more abstract and disembodied interactions. We start by focusing on how early infant-caregiver bodily interactions play a crucial role in shaping the boundaries of the self but also in learning to predict others' behavior. We then explore the social function of the sense of touch in the entire life span, highlighting its role in promoting physical and psychological well-being by supporting positive interpersonal exchanges. We go on by introducing the concept of implicit theory of mind, as the early ability to interpret others' intentions, possibly grounded in infant-caregiver bodily exchanges (embodied practices). In the following part, we consider so-called higher level forms of social interaction: intellectual exchanges among individuals. In this regard, we defend the view that, beside the apparent private dimension of "thinking abstractly", using abstract concepts is intrinsically a social process, as it entails the re-enactment of the internalized dialogue through which we acquired the concepts in the first place. Finally, we describe how the hypothesis of "dialectical attunement" may explain the development of abstract thinking: to effectively transform the world according to their survival needs, individuals co-construct structured concepts of it; by doing so, humans fundamentally transform not merely the world they are being in, but their being in the world.

The use of gaze to study cognition: limitations, solutions, and applications to animal welfare

The study of gaze responses, typically using looking time paradigms, has become a popular approach to improving our understanding of cognitive processes in non-verbal individuals. Our interpretation of data derived from these paradigms, however, is constrained by how we conceptually and methodologically approach these problems. In this perspective paper, we outline the application of gaze studies in comparative cognitive and behavioral research and highlight current limitations in the interpretation of commonly used paradigms. Further, we propose potential solutions, including improvements to current experimental approaches, as well as broad-scale benefits of technology and collaboration. Finally, we outline the potential benefits of studying gaze responses from an animal welfare perspective. We advocate the implementation of these proposals across the field of animal behavior and cognition to aid experimental validity, and further advance our knowledge on a variety of cognitive processes and welfare outcomes.

Socially evaluative contexts facilitate mentalizing

Our ability to understand others' minds stands at the foundation of human learning, communication, cooperation, and social life more broadly. Although humans' ability to mentalize has been well-studied throughout the cognitive sciences, little attention has been paid to whether and how mentalizing differs across contexts. Classic developmental studies have examined mentalizing within minimally social contexts, in which a single agent seeks a neutral inanimate object. Such object-directed acts may be common, but they are typically consequential only to the object-seeking agent themselves. Here, we review a host of indirect evidence suggesting that contexts providing the opportunity to evaluate prospective social partners may facilitate mentalizing across development. Our article calls on cognitive scientists to study mentalizing in contexts where it counts.

Theory of mind tested by implicit false belief: a simple and full-fledged mental state attribution

About 40 years have passed since 'theory of mind (ToM)' research started. The false-belief test is used as a litmus test for ToM ability. The implicit false-belief test has renewed views of ToM in several disciplines, including psychology, psychiatry, and neuroscience. Many important questions have been considered via the paradigm of implicit false belief. We recently addressed the phylogenetic and physiological aspects of ToM using a version of this paradigm combined with the chemogenetic technique on Old World monkeys. We sought to create animal models for autism that exhibit behavioral phenotypes similar to human symptoms. The simultaneous manipulation of neural circuits and assessments of changes in phenotypes can help identify the causal neural substrate of ToM.

Do young children track other's beliefs, or merely their perceptual access? An interactive, anticipatory measure of early theory of mind

This paper aimed to contribute to answering three questions. First, how robust and reliable are early implicit measures of false belief (FB) understanding? Second, do these measures tap FB understanding rather than simpler processes such as tracking the protagonist's perceptual access? Third, do implicit FB tasks tap an earlier, more basic form of theory of mind (ToM) than standard verbal tasks? We conducted a conceptual replication of Garnham & Perner's task (Garnham and Perner 2001 Br. J. Dev. Psychol. 19, 413-432) simultaneously measuring children's anticipatory looking and interactive behaviours toward an agent with a true or FB (N = 81, M = 40 months). Additionally, we implemented an ignorance condition and a standard FB task. We successfully replicated the original findings: children's looking and interactive behaviour differed according to the agent's true or FB. However, children mostly did not differentiate between FB and ignorance conditions in various measures of anticipation and uncertainty, suggesting the use of simpler conceptual strategies than full-blown ToM. Moreover, implicit measures were all related to each other but largely not related to performance in the standard FB task, except for first look in the FB condition. Overall, our findings suggest that these implicit measures are robust but may not tap the same underlying cognitive capacity as explicit FB tasks.

Evidence for goal- and mixed evidence for false belief-based action prediction in two- to four-year-old children: A large-scale longitudinal anticipatory looking replication study

Unsuccessful replication attempts of paradigms assessing children's implicit tracking of false beliefs have instigated the debate on whether or not children have an implicit understanding of false beliefs before the age of four. A novel multi-trial anticipatory looking false belief paradigm yielded evidence of implicit false belief reasoning in three- to four-year-old children using a combined score of two false belief conditions (Grosse Wiesmann, C., Friederici, A. D., Singer, T., & Steinbeis, N. [2017]. Developmental Science, 20(5), e12445). The present study is a large-scale replication attempt of this paradigm. The task was administered three times to the same sample of N = 185 children at two, three, and four years of age. Using the original stimuli, we did not replicate the original finding of above-chance belief-congruent looking in a combined score of two false belief conditions in either of the three age groups. Interestingly, the overall pattern of results was comparable to the original study. Post-hoc analyses revealed, however, that children performed above chance in one false belief condition (FB1) and below chance in the other false belief condition (FB2), thus yielding mixed evidence of children's false belief-based action predictions. Similar to the original study, participants' performance did not change with age and was not related to children's general language skills. This study demonstrates the importance of large-scaled replications and adds to the growing number of research questioning the validity and reliability of anticipatory looking false belief paradigms as a robust measure of children's implicit tracking of beliefs. This article is protected by copyright. All rights reserved.